Qwen 3.5 2B (Non-reasoning): benchmark results
Qwen 3.5 2B evaluated with reasoning disabled. Provider: Alibaba. Released 2026-03-02. Access: Open.
Aggregate Rating 1354 ± 15, rank #1678 of 2168 rated models, from 104 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| UnpredictaBench - Shuffling | 36.84 | KS@100 (%): share of the 19 Shuffling problems (random order | 92.1 |
| UnpredictaBench - RealWorld | 31.03 | KS@100 (%): share of the 30 RealWorld problems (natural-lang | 78.9 |
| PRISM-VLM - Counting | 57.9 | Per-axis mean score (%; exact-integer accuracy on synthesize | 75.6 |
| UnpredictaBench - Code | 12.61 | KS@100 (%): share of the 238 Code problems (distributions in | 73.7 |
| PRISM-VLM - Sycophancy Resistance | 65.6 | Per-axis mean score (%; share of turn-0-correct answers held | 73.2 |
| Tau2-Bench Telecom | 81.58 | Success Rate (%) | 72.8 |
| AA TAU-2 Bench | 81.58 | Accuracy (%) | 72.3 |
| UnpredictaBench - Text | 14.38 | KS@100 (%): share of the 160 Text problems (named or describ | 63.2 |
| PRISM-VLM | 54.9 | PScore (%; harmonic mean of the seven per-axis mean scores, | 60 |
| AA Omniscience - Software Engineering (SWE) - Julia | 12 | Accuracy (%) | 58 |
| PRISM-VLM - Hallucination Resistance | 59.3 | Per-axis mean score (%; share of trap questions (absent enti | 56.1 |
| UGI - Willingness (W/10) | 6 | W/10 Score | 55.1 |
Interactive version: theaggregate.ai/model?slug=qwen-3-5-2b-non-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.